AIPOOOL Weekly AI Digest – 13 July 2025

Happy Sunday! This is AIPOOOL. The email that tells you what’s going on in Artificial Intelligence space in simple blocks. Get ready to have your mind blown by the sheer power of AI!

In Today’s Email :

• 🔥AI News Flash: McDonald's AI hiring tool exposes 64M applicants in a massive data breach, Goldman Sachs pilots an autonomous AI coder, Microsoft unveils a model to supercharge drug discovery, Perplexity launches its AI-powered Comet browser, and Moonshot AI releases its powerful open-source Kimi K2 model!

• ⛏️ Trending Tools: PERSO.ai for AI Dubbing, Inkr for Transcribe & many more …

• 🔰 Quick Grab: Simular's Agent S2 Sets New Standard for AI Computer Interaction

• 🎆Creators Corner: Top Picks from Hugging Face: Trending AI Applications You Can't Miss!

• 🥼 From Lab to Layman: EraRAG: Revolutionizing RAG for Dynamic Knowledge Environments

Browse AI Tools | Instagram | Advertise

AI Happenings You Don’t Want To Miss

✨ McDonald's AI Hiring Tool Exposes Applicant Data
A significant data breach at McDonald's exposed the information of approximately 64 million job applicants due to a laughably simple password, "123456," on its AI-powered hiring platform. This incident highlights the critical need for stronger security measures, even with third-party vendors.

✨ China's Moonshot AI Releases Open-Source Kimi K2 Model
Seeking to reclaim its market position, Chinese AI startup Moonshot AI has launched Kimi K2, a powerful open-source model. This release is part of a growing trend among Chinese tech firms to open-source their AI models to foster innovation and adoption.

✨ Goldman Sachs Pilots Autonomous AI Coder
In a major milestone for Wall Street, Goldman Sachs is testing an autonomous AI software engineer named Devin, created by the startup Cognition. This advanced AI can handle complex coding tasks independently, potentially boosting developer productivity by three to four times.

✨ Perplexity Launches Comet, an AI-Powered Browser
Perplexity has introduced its first AI-powered web browser, Comet, designed to compete with Google Chrome. Comet integrates Perplexity's AI search capabilities and an AI assistant to summarize content, manage tabs, and automate web navigation.

✨ Microsoft Unveils BioEmu-1 for Accelerated Drug Discovery
Microsoft has introduced BioEmu-1, a deep learning model that can generate thousands of protein structures per hour on a single GPU. This innovation is 10,000 to 100,000 times more efficient than traditional methods, significantly accelerating drug discovery and protein engineering.

✨ Moonshot AI Launches Agentic Model Kimi K2
Moonshot AI has released Kimi K2, an open-source model with one trillion total parameters, specifically optimized for coding and agentic tasks. The model demonstrates state-of-the-art performance on several key benchmarks, outperforming many existing open models.

Free & Useful AI Tools -

1. PERSO.ai- AI dubbing: Create global videos instantly.

2. Model Playground AI - Compare AI models side by side, hassle-free.

3. Inkr - Transcribe audio and video effortlessly in minutes.

4. SJinn - AI-powered content creation from text to visuals.

📜 Simular's Agent S2 Sets New Standard for AI Computer Interaction

Simular AI has released Agent S2, an open-source framework that marks a significant leap forward for autonomous AI agents that interact with graphical user interfaces (GUIs). By employing a novel compositional generalist-specialist architecture, Agent S2 effectively delegates tasks between high-level planning models and specialized modules for precise execution, overcoming the limitations of single-model approaches. This allows the agent to handle complex, multi-step workflows across various applications and operating systems with greater accuracy.

The framework's advanced design has led to state-of-the-art performance on major benchmarks, outperforming previous leaders on OSWorld and AndroidWorld. Agent S2’s ability to learn from past interactions and continually improve its knowledge base sets it apart. It integrates tools like UI-TARS for visual grounding and Perplexica for web-knowledge retrieval, enabling it to robustly handle tasks requiring both GUI navigation and real-time information. Agent S2 provides a powerful infrastructure for developers building more reliable and production-ready automation, QA, and accessibility solutions.

🤖 Top Picks from Hugging Face: Trending AI Applications You Can't Miss!

✨ FastVLM on WebGPU: High-Speed Vision-Language Model
This space from the WebML community showcases a vision-language model running efficiently in a web browser using WebGPU, enabling fast and interactive object detection from text prompts.

✨ Kontext Relight: Advanced Image Lighting Control
A tool from the kontext-community that allows users to realistically relight images by adding, modifying, and removing light sources with impressive control and quality.

✨ SpatialTrackerV2: Advanced Video Object Tracking
This interactive space by Yuxihenry demonstrates a powerful model for tracking and segmenting any object within a video, allowing users to define points of interest for precise monitoring.

✨ All-Tracker: Universal Object Tracking and Segmentation
A versatile tool by aharley that can track and segment virtually any object in a video. Users can simply click to define a target, and the model will follow it through the footage.

👨‍💻 EraRAG: Revolutionizing RAG for Dynamic Knowledge Environments

Researchers have introduced EraRAG, a novel framework designed to address a critical limitation in traditional Retrieval-Augmented Generation (RAG) systems: their inefficiency when dealing with continuously expanding datasets. Standard RAG models require costly and time-consuming full re-indexing and retraining whenever new documents are added to their knowledge corpus.

EraRAG overcomes this challenge with an efficient and incremental update mechanism. The framework features an incrementally updatable retriever that seamlessly integrates new documents without rebuilding the entire index from scratch. Complementing this is a continually trained generator that adapts to the expanding knowledge base by learning from new data without forgetting previously acquired information.

This innovative approach significantly reduces the computational overhead and time required to keep AI models current with the latest information. By enabling RAG systems to evolve dynamically with growing corpora, EraRAG represents a crucial advancement for developing practical, scalable, and perpetually up-to-date knowledge-intensive applications.

We’re Curious… What we should cover more?

Click HERE to provide your feedback.

Do us a favor? Reply to this email and tell us what you'd like to see more (or less) of!

How did we do?

Click HERE to provide your feedback.